In [111]:
import json
import pathlib
from pprint import pprint
import matplotlib.pyplot as plt
import polars as pl
import srcdir
BBOX_IMAGE_DIR = pathlib.Path('../data/train_bbox_images')
BBOX_ANNOTATIONS = pathlib.Path('../data/train_bbox_annotations.json')
In [87]:
with open(BBOX_ANNOTATIONS, 'r') as f:
annotations = json.load(f)['images']
bboxes = []
for annotation in annotations:
_anno = {
'image_id': int(annotation['file_name'].split('.')[0].split('_')[-1]),
'file_name': annotation['file_name'],
'image_width': annotation['width'],
'image_height': annotation['height'],
}
for i, bbox in enumerate(annotation['annotations']):
_bbox = {
'bbox_id': i,
'class': bbox['class'],
'x_min': bbox['bbox'][0],
'y_min': bbox['bbox'][1],
'bbox_width': bbox['bbox'][2],
'bbox_height': bbox['bbox'][3],
'bbox_area': bbox['bbox'][2] * bbox['bbox'][3],
}
bboxes.append(_anno | _bbox)
bbox_df = pl.DataFrame(bboxes).sort(by=('image_id', 'bbox_id'))
bbox_df
Out[87]:
shape: (2_653, 11)
| image_id | file_name | image_width | image_height | bbox_id | class | x_min | y_min | bbox_width | bbox_height | bbox_area |
|---|---|---|---|---|---|---|---|---|---|---|
| i64 | str | i64 | i64 | i64 | str | f64 | f64 | f64 | f64 | f64 |
| 0 | "train_0.tif" | 500 | 375 | 0 | "vacant_lot" | 311.0 | 36.246 | 35.0 | 45.146 | 1580.11 |
| 0 | "train_0.tif" | 500 | 375 | 1 | "vacant_lot" | 121.0 | 156.0 | 203.0 | 144.0 | 29232.0 |
| 1 | "train_1.tif" | 500 | 375 | 0 | "vacant_lot" | 198.0 | 231.0 | 56.0 | 34.0 | 1904.0 |
| 2 | "train_2.tif" | 500 | 375 | 0 | "vacant_lot" | 61.501 | 168.404 | 47.705 | 45.298 | 2160.94109 |
| 3 | "train_3.tif" | 500 | 375 | 0 | "vacant_lot" | 327.0 | 196.0 | 33.0 | 57.0 | 1881.0 |
| … | … | … | … | … | … | … | … | … | … | … |
| 598 | "train_598.tif" | 500 | 375 | 11 | "vacant_lot" | 470.0 | 162.0 | 27.0 | 28.0 | 756.0 |
| 599 | "train_599.tif" | 500 | 375 | 0 | "vacant_lot" | 71.509 | 259.731 | 53.675 | 67.836 | 3641.0973 |
| 599 | "train_599.tif" | 500 | 375 | 1 | "vacant_lot" | 223.0 | 136.0 | 44.0 | 44.0 | 1936.0 |
| 599 | "train_599.tif" | 500 | 375 | 2 | "vacant_lot" | 217.0 | 185.0 | 43.0 | 45.0 | 1935.0 |
| 599 | "train_599.tif" | 500 | 375 | 3 | "vacant_lot" | 66.0 | 215.0 | 32.0 | 45.0 | 1440.0 |
In [88]:
print('Number of images:', len(annotations))
print('Number of bboxes:', len(bbox_df))
Number of images: 600 Number of bboxes: 2653
In [89]:
bbox_df.get_column('image_width').unique()
Out[89]:
shape: (1,)
| image_width |
|---|
| i64 |
| 500 |
In [91]:
bbox_df.get_column('image_height').unique()
Out[91]:
shape: (1,)
| image_height |
|---|
| i64 |
| 375 |
In [92]:
bbox_df.get_column('class').unique()
Out[92]:
shape: (1,)
| class |
|---|
| str |
| "vacant_lot" |
In [93]:
bbox_counts = bbox_df.group_by('image_id') \
.agg(n_bboxes=pl.col('bbox_id').max() + 1) \
.get_column('n_bboxes') \
.value_counts() \
.sort(by='n_bboxes')
plt.bar(
bbox_counts.get_column('n_bboxes').to_numpy(),
bbox_counts.get_column('count').to_numpy(),
)
plt.xlabel('Number of bboxes per image')
plt.ylabel('Number of images')
bbox_counts
Out[93]:
shape: (25, 2)
| n_bboxes | count |
|---|---|
| i64 | u32 |
| 1 | 128 |
| 2 | 114 |
| 3 | 96 |
| 4 | 68 |
| 5 | 36 |
| … | … |
| 21 | 1 |
| 22 | 1 |
| 24 | 2 |
| 27 | 1 |
| 30 | 1 |
In [94]:
bbox_df.get_column('bbox_width').describe(percentiles=None)
Out[94]:
shape: (6, 2)
| statistic | value |
|---|---|
| str | f64 |
| "count" | 2653.0 |
| "null_count" | 0.0 |
| "mean" | 59.62923 |
| "std" | 38.379945 |
| "min" | 5.0 |
| "max" | 469.0 |
In [95]:
bbox_df.get_column('bbox_height').describe(percentiles=None)
Out[95]:
shape: (6, 2)
| statistic | value |
|---|---|
| str | f64 |
| "count" | 2653.0 |
| "null_count" | 0.0 |
| "mean" | 55.043907 |
| "std" | 33.522681 |
| "min" | 1.0 |
| "max" | 375.0 |
In [96]:
bbox_df.get_column('bbox_area').describe(percentiles=None)
Out[96]:
shape: (6, 2)
| statistic | value |
|---|---|
| str | f64 |
| "count" | 2653.0 |
| "null_count" | 0.0 |
| "mean" | 4116.176349 |
| "std" | 6549.882811 |
| "min" | 14.0 |
| "max" | 175406.0 |
In [98]:
bbox_df.filter(
(pl.col('bbox_height') == bbox_df.get_column('bbox_height').min())
| (pl.col('bbox_height') == bbox_df.get_column('bbox_height').max())
| (pl.col('bbox_width') == bbox_df.get_column('bbox_width').min())
| (pl.col('bbox_width') == bbox_df.get_column('bbox_width').max())
| (pl.col('bbox_area') == bbox_df.get_column('bbox_area').min())
| (pl.col('bbox_area') == bbox_df.get_column('bbox_area').max())
| (pl.col('bbox_id') == bbox_df.get_column('bbox_id').max())
)
Out[98]:
shape: (5, 11)
| image_id | file_name | image_width | image_height | bbox_id | class | x_min | y_min | bbox_width | bbox_height | bbox_area |
|---|---|---|---|---|---|---|---|---|---|---|
| i64 | str | i64 | i64 | i64 | str | f64 | f64 | f64 | f64 | f64 |
| 30 | "train_30.tif" | 500 | 375 | 10 | "vacant_lot" | 141.0 | 374.0 | 14.0 | 1.0 | 14.0 |
| 79 | "train_79.tif" | 500 | 375 | 0 | "vacant_lot" | 0.0 | 1.0 | 469.0 | 374.0 | 175406.0 |
| 238 | "train_238.tif" | 500 | 375 | 6 | "vacant_lot" | 0.0 | 212.0 | 5.0 | 15.0 | 75.0 |
| 435 | "train_435.tif" | 500 | 375 | 1 | "vacant_lot" | 0.0 | 0.0 | 91.0 | 375.0 | 34125.0 |
| 453 | "train_453.tif" | 500 | 375 | 29 | "vacant_lot" | 261.0 | 65.0 | 53.0 | 48.0 | 2544.0 |
In [129]:
def get_bboxes(
df: pl.DataFrame,
image_id: int,
bbox_ids: int | list[int] | None = None
) -> tuple[int, int, int, int]:
if bbox_ids is None:
annotations = df.filter(
(pl.col('image_id') == image_id)
)
else:
if not hasattr(bbox_ids, '__iter__'):
bbox_ids = [bbox_ids]
annotations = df.filter(
(pl.col('image_id') == image_id) & (pl.col('bbox_id').is_in(bbox_ids))
)
bboxes = annotations.select('x_min', 'y_min', 'bbox_width', 'bbox_height') \
.to_numpy() \
.tolist()
return tuple(bboxes)
def show_bboxes(imgfile: pathlib.Path, bboxes: tuple[int] | list[tuple[int]]) -> None:
img = plt.imread(imgfile)
plt.figure(figsize=(10, 10))
plt.imshow(img)
for bbox in bboxes:
plt.gca().add_patch(
plt.Rectangle(
(bbox[0], bbox[1]), bbox[2], bbox[3],
fill=False, edgecolor='red', linewidth=1
)
)
plt.axis('off')
plt.show()
In [130]:
bboxes = get_bboxes(bbox_df, image_id=30, bbox_ids=10)
pprint(bboxes)
show_bboxes(
imgfile=BBOX_IMAGE_DIR / 'train_30.tif',
bboxes=bboxes
)
([141.0, 374.0, 14.0, 1.0],)
In [131]:
bboxes = get_bboxes(bbox_df, image_id=79, bbox_ids=0)
pprint(bboxes)
show_bboxes(
imgfile=BBOX_IMAGE_DIR / 'train_79.tif',
bboxes=bboxes
)
([0.0, 1.0, 469.0, 374.0],)
In [133]:
bboxes = get_bboxes(bbox_df, image_id=238, bbox_ids=6)
pprint(bboxes)
show_bboxes(
imgfile=BBOX_IMAGE_DIR / 'train_238.tif',
bboxes=bboxes
)
([0.0, 212.0, 5.0, 15.0],)
In [134]:
bboxes = get_bboxes(bbox_df, image_id=435, bbox_ids=1)
pprint(bboxes)
show_bboxes(
imgfile=BBOX_IMAGE_DIR / 'train_435.tif',
bboxes=bboxes
)
([0.0, 0.0, 91.0, 375.0],)
In [135]:
bboxes = get_bboxes(bbox_df, image_id=453, bbox_ids=None)
pprint(bboxes)
show_bboxes(
imgfile=BBOX_IMAGE_DIR / 'train_453.tif',
bboxes=bboxes
)
([89.742, 217.071, 45.586, 35.236999999999995], [121.0, 315.0, 39.0, 44.0], [121.0, 340.0, 17.0, 16.0], [156.0, 142.0, 72.0, 58.0], [138.0, 177.0, 41.0, 32.0], [100.0, 160.0, 39.0, 47.0], [28.0, 131.0, 38.0, 60.0], [25.0, 187.0, 30.0, 22.0], [21.0, 212.0, 32.0, 30.0], [19.0, 243.0, 33.0, 31.0], [14.0, 273.0, 36.0, 31.0], [11.0, 303.0, 35.0, 29.0], [35.0, 336.0, 27.0, 32.0], [9.0, 333.0, 43.0, 29.0], [54.0, 209.0, 17.0, 30.0], [53.0, 238.0, 16.0, 27.0], [50.0, 265.0, 17.0, 29.0], [47.0, 294.0, 18.0, 26.0], [46.0, 321.0, 16.0, 15.0], [223.0, 170.0, 54.0, 44.0], [219.0, 214.0, 39.0, 35.0], [195.0, 207.0, 33.0, 66.0], [224.0, 214.0, 35.0, 35.0], [215.0, 247.0, 37.0, 34.0], [189.0, 275.0, 46.0, 43.0], [271.0, 305.0, 48.0, 38.0], [80.0, 316.0, 39.0, 38.0], [257.0, 321.0, 71.0, 54.0], [277.0, 221.0, 51.0, 74.0], [261.0, 65.0, 53.0, 48.0])
